Pages

Thursday, June 16, 2016

Cognitive Computing – fitting the platform to the application

By Rich Ptak


Recently, it has been popular to assert that IT professionals and business staff need no longer concern
themselves or consider IT infrastructure. Among other reasons, the claim is that with the growth of Cloud computing and commoditization, infrastructure no longer matters. The assertion being that a general purpose architecture provides all the processing flexibility and power needed to deliver a range of services.

We are convinced that this view is short-sighted and wrong as it ignores the changing dynamics of computing as Moore’s Law runs down, and technology evolves. It focuses on traditional performance metrics while ignoring the realities disrupting IT, how it is designed, implemented and realized in operations and applications.
It trivializes the difficulties in evolving and delivering systems that are able to meet the requirements of the high speed, data-intensive, highly scalable, adaptable workloads associated with evolving technologies of genomics, nanotechnology, etc. It denies or wishes to ignore the need for and interest in open, standards-based systems-oriented infrastructure able to intelligently adapt and optimize for evolving workloads. 

Identifying the Future of Infrastructure

There are enterprise and IT professionals who recognize and understand the implications of the extraordinary demands placed on infrastructure as a result of the combination of today’s competitive market and evolving technologies.

With the IBM IT Infrastructure Point of View[1] (POV) website, IBM offers these professionals “thought leadership” opinions and insights that promote and support the role of IT leaders in planning the use of technology to achieve organizational success. The target audience are those who not only understand the demand, but also seek to add to their knowledge in order to better prepare themselves and their organizations to meet future challenges.

Today’s Digital enterprises are being challenged to meet escalating expectations of clients and customers for extraordinary performance, dynamic scalability, robust adaptability and rapid innovation in the delivery of services and products. Such demands cannot be met with infrastructure and systems compromised to provide common denominator needs. Nor, can it be met with the ‘static’ configurations and fixed architectures of the very recent past.

Meeting the evolving demands of data and compute intensive digital enterprise, requires a systems infrastructure for server and storage operations that can be intelligently optimized. The infrastructure must be optimized to deal with emerging, evolving styles of computing. It must be flexible enough to integrate and interact with emerging technologies while still able to interoperate with existing operating environments. It must also be intelligently and cognitively adaptable to meet the emerging demands of whatever workload it takes on.  

IBM’s Point of View on IT Infrastructure for Cognitive Workloads

With the explosion of Cognitive computing, its hybrid cloud platforms, Power and z Systems, storage solutions and experience in leading edge technologies and solutions, IBM is uniquely positioned to work with clients to help to shape the future of their computing operations. Enterprise IT must not only get the most from their existing infrastructure but must also act to leverage new cognitive capabilities and take advantage of emerging technologies.

Today’s systems-oriented solutions depend upon server and storage technologies that can be combined with software driven cognitive abilities, such as IBM’s Watson. Cognitive computing has the potential to understand, reason, learn and adapt to changes in their operational environment and workloads. IBM is working with clients, customers and partners to push the boundaries of what is possible with an IT infrastructure optimized for cognitive workloads. 

In recognition of all this, IBM is publishing a Point of View (PoV) about IT infrastructure and cognitive workloads. It describes in significant detail how IBM, in conjunction with its partners, will advise and work with customers to aid them in their efforts to organize and plan in order to gain the most advantage from their IT systems and storage infrastructure.

As would be expected, key to this approach are the IBM solutions portfolio of z Systems, Power Systems, IBM Storage, hybrid cloud services and software-driven cognitive computing (ala Watson).  
This strategy is built around three principles:   
  1.     Design for Cognitive Business – to allow action at the speed of thought,
  2.    Build with Collaborative Innovation – to accelerate technology breakthroughs,
  3.     Delivery through a Cloud Platform – to extend the value of systems and data. 
IBM has multiple projects (both under-way and completed) where cognitive computing has provided the key factor in achieving competitive advantage, financial performance and enterprise success. The involve enterprises and organizations in a wide variety of markets. The projects have accelerated time to insights with infrastructure deliberately designed and architected for unstructured data with companies in banking, oil and gas exploration, and academia. They have sped-up development of new solutions while cutting the time-to-market. They have provided infrastructure optimized to run specific workloads with unique business requirements for customers ranging from government agencies to healthcare services.

We won’t steal any more of IBM’s thunder. We suggest that you visit the IBM IT infrastructure site to review the details. In our opinion, IBM appears to be well ahead of its competition with its comprehensive, customer-centric view and vision. They are also uniquely positioned to speak on this topic. They have both the cutting edge technology and significant real-life implementation experience with products to demonstrate their ability to deliver on their visions for the future of cognitive computing supported by intelligent infrastructure.

Tuesday, June 14, 2016

Acceleration, Collaboration, Innovation - IBM's roadmap for its POWER architecture

By Bill Moran and Rich Ptak


Vendors know the wisdom of publishing a product roadmap. Users want to know the planned future
of the products that they might invest in. They also want insight into how the vendor sees the product evolving.

So, IBM has reason to present the POWER architecture’s future to potential customers and partners. Having successfully persuaded many companies to sign up for OpenPOWER systems IBM must address questions concerning the future of the product’s architecture. IBM laid out its architectural strategy along with some specifics on its future. We discuss key takeaways from that presentation.



NVLink will be available in systems later this year and be carried forward in future systems. Notice that NVLink and CAPI are both specialized technologies for boosting certain kinds of performance. Combined with various architectural changes, they compensate for the rundown of Moore’s Law. Expect to see more such technologies in the future.

The current POWER8 system architecture is based on a 22 nm chip with 12 cores. Announced in 2014, IBM plans to continue with that base until mid-2017. The major enhancement to current version is the addition of NVIDIA NVLink. This acts as an extremely high –speed interconnect link between the chip and an NVIDIA GPU. The link delivers 80 GB per second in each direction. This is 5 to 12 times faster than today’s fastest available link.  The NVIDIA GPU accelerates floating point operations and other numerically intensive operations that are common in cognitive computing, analytics, and high performance computing.



Featuring partner-developed microprocessors in a roadmap is unique to IBM. It dramatically underscores the vitality of OpenPOWER activities. To our knowledge, no other hardware vendor has achieved anything like this!

IBM and its partners will build systems to take maximum advantage of this link which allows parallel processing using NVIDIA GPU's. IBM identified two additional partners, Zoom Netcom and Inventec Corporation. Zoom is a China-based system board developer. Inventec is a Taiwan-based server and laptop company. We expect both of these companies to be working on systems for their focus areas.

In mid-2017, IBM will begin rolling out Power9, a 14 nm chip versus today’s 22 nm Power8. IBM will first introduce a 24 core scale-out system. This will be followed sometime later with scale-up versions. There was no statement on the number of cores in scale up systems. The POWER9 systems will feature a new micro-architecture built around 24 newly redesigned cores and including a number of high-speed cache and memory interconnects including DDR4 direct attach memory channels, PCIe gen4 and custom accelerators from IBM and its partners.

In the time period between 2018 and 2019 IBM expects its partners to announce chip offerings based on IP from both Power8 and Power9, based on 10 to 7 nm technology. Partners will be targeting offerings to their own specialized market segments.

While IBM avoided any claims, we expect these systems’ shrinking chip technology will have some dramatic effects (upward) on demand. It’s also clear the partners expect to gain significant competitive and business advantages from their efforts.

Power10, expected sometime after 2020, is the next large step into the future. IBM provided no details on features or performance. Typical for a product at least 4 years away.


IBM will offer two Power9 families. Initially the focus will be scale-out systems with a maximum of 24 cores. Later, scale-up systems will be added, presumably with a larger number of cores. They will share a common architecture.

This road map shows that IBM along with other members of the OpenPOWER Foundation are developing Power at an increasing rate. Remember, IBM’s POWER group faces a number of unique and new challenges. No other major vendor has ever attempted to develop new hardware in collaboration with numerous partners ala the OpenPOWER Foundation.

Also, since selling its chip production facilities to Global Foundries, IBM’s POWER people must negotiate with an outside company. We believe that IBM’s POWER team are doing an exceptional job in coping with these difficulties. If they deliver on the items in this road map, the architecture should remain competitive. Chips developed by other companies provide both a roadmap highlight and effectively demonstrate Power Foundation’s strength.

Here is our simplified version of the road-map (with acknowledgement to IBM):

        Today                2H 2016               2017                   TBD                2018-2019           2020+
Power8
Power8
Power9
Power9
Power8/9
Power10
12 cores
12 cores
24 cores
? cores


CAPI
NVLINK + CAPI
Scale out
New Arch
Scale up
Partner developed.

22 nm
22 nm
14 nm
14 nm
10-7 nm